Rohan Kottamasu

| Founder: @EnigmaProfits @ParagonAIO @ParagonNotify |
Currently Building ML Pipelines - Deep Variance

[Rohan Kottamasu @VertexAlts]

Accomplishments :

[Rohan Kottamasu @VertexAlts]

We’re building an optimization later in the AI training stack. We want to make AI affordable

We’ve developed research that optimizes memory allocation in GPUs, so we effectively lower compute by 24% on avg and increase throughout by 32%.

We’ve productionized SDKs, and now offer products that optimize both GPUs and LLMs.

[Rohan Kottamasu @VertexAlts]

Saw your field notes on scaling MoE expert parallelism to 128 GPUs. Wild how memory utilization per GPU drops off so much at that size.

With the clusters you guys run I bet squeezing more effective compute out of every GPU would be a huge win for training speed and cost.

My team has research that fixes exactly that kind of hidden memory waste at the lowest level with zero code changes. Curious if this is still biting you.
@xariusrke @NousResearch

[Nous Research @NousResearch]

Follow along with researcher @xariusrke in these detailed field notes as he hunts down MoE training bottlenecks

https://t.co/0bVb2Sb0bf

[Girish @girish327]

Today we open-sourced deep_variance, a Python SDK designed to reduce GPU memory overhead and improve efficiency in deep learning training.

Hi All,
I worked on this open-source SDK package together with Saai Vignesh Premanand as part of the Deep Variance initiative.
The goal of this SDK is to help researchers and engineers run large deep learning training more efficiently by improving GPU memory usage and enabling more scalable experimentation without constantly running into memory limits.

You can install it directly from PyPI and start using it in your deep learning workflows.

🔗 https://t.co/jDFnQypX89

We hope this helps developers and researchers working on deep learning systems. Looking forward to feedback from the community.

FYI: The beta version is for any NVIDIA GPU with CUDA and C++ environment.

@nvidia @NVIDIAAI @NVIDIAGeForce @Google @googledevs #opensource #GPU #CUDA @GoogleDeepMind @sundarpichai @AMD @intel @ionet @akashnet @Togetherdec @digitalocean @awscloud @AWSstartups @GoogleStartups @googlecloud @Azure @Microsoft

[Rohan Kottamasu @VertexAlts]

@FurqanR @NebulaAI At Deep Variance we focus on research to optimize GPUs. We are reducing clock time by at least 25%.

Lets Chat!

[Rohan Kottamasu @VertexAlts]

@girish327 @fdotinc We’d use funding towards accelerating our research (compute), development and GTM (ad-spend).

@fdotinc

[Enigma @EnigmaProfits]

Free Lifetime Giveaway 🥳

Rules :
~ Follow @LucrorIO
~ Tag 1 Friend
~ Retweet ♻️

---

With our recent Stellar Investment Call,
members were able to double their money over a period of couple days! 🤭

[Enigma @EnigmaProfits]

Free Lifetime Giveaway 🥳

Rules :
~ Follow @LucrorIO
~ Tag 1 Friend
~ Retweet ♻️

---

This past month, users made tons of money off investing!
Excluding the profits from restocks, users were given the opportunity to make over $1.7k+ from our calls alone!

[Rohan Kottamasu @VertexAlts]

My team is building a research lab that democratizes access to serious predictive ML, deep learning, and LLM fine tuning. Today you need expensive infra + an ML team.

We’re changing that. We ship dev tools that:
• Automate the full model building pipeline (making it easy to use)
• Cut compute & time to ship by >40% vs typical AutoML at similar accuracy

We're looking for investment ASAP to scale fast & get access to infrastructure to help us build. When we scale, every business runs on enterprise level ML

[Rohan Kottamasu @VertexAlts]

My team is building a research lab that democratizes access to serious predictive ML, deep learning, and LLM fine tuning.

Today you need expensive infra + an ML team.
We’re changing that. We ship dev tools that:
• Automate the full model building pipeline (making it easy to use)
• Cut compute & time to ship by >40% vs typical AutoML at similar accuracy

When we scale, every business runs on enterprise level ML

[Rohan Kottamasu @VertexAlts]

I’m building a research lab that democratizes access to serious predictive ML and deep learning. Today you need expensive infra + an ML team.

We’re changing that. We ship dev tools that:
• Automate the full ML + deep learning pipeline (making it easy to use)
• Cut compute by >40% vs typical AutoML at similar accuracy

When this works, every business runs on production ML

[Rohan Kottamasu @VertexAlts]

@PeterDAmbrosio Hi Peter, would appreciate a list of grants

[Rohan Kottamasu @VertexAlts]

@s_chiriac Research Lab Democratizing ML & Deep-Learning
https://t.co/xrwsEuzJMa

[Rohan Kottamasu @VertexAlts]

I’m building a research lab that democratizes access to serious predictive ML and deep learning.
Today you need expensive infra + an ML team. For that rzn, trillions get poured into AI with little tangible business value. We’re changing that.

We ship dev tools that:
• Auto-configure the full ML + deep learning pipeline (preprocessing, model family, architecture, hyperparams)
• Learn deep-net architectures and training schedules for the task at hand
• Cut compute by ~70% vs typical AutoML at similar accuracy

The vision is to stay a research lab that always ships products backed by our own research, so any business can use ML and DL in its decision-making, no matter the constraints.

When this works, every business runs on production ML

[Rohan Kottamasu @VertexAlts]

@davefontenot this guy @shreshthrajan

[Rohan Kottamasu @VertexAlts]

@KanyeBrother Don't be sad that it's over.
Smile because it happened.

[Rohan Kottamasu @VertexAlts]

@bkantha1 @ParagonAIO ?

[Rohan Kottamasu @VertexAlts]

@AceFarming Legit. Smooth Transaction

[Rohan Kottamasu @VertexAlts]

@umarz__ Legit. Smooth Transaction

[Rohan Kottamasu @VertexAlts]

@notsocratese Another competitor ...
Don't forget, @ParagonAIO coming for that crown too.

[Rohan Kottamasu @VertexAlts]

@SuspendedAndy Those are just discouraging words to new botters out there.

More so, do your research into the owners / company / people involved before you buy into something.